| 1 |
On 05/07/2012 12:33 PM, Stelian Ionescu wrote: |
| 2 |
> Isn't it the time to make a new EAPI which no longer has USE "flags" but |
| 3 |
> USE "values" ? Many of the really weird USE flags combinations would be |
| 4 |
> much more clearly expressed if the possible types for a USE variable |
| 5 |
> were: |
| 6 |
> 1) member-of: for choosing the backend of certain functionality; e.g. |
| 7 |
> all combinations of USE="ssl openssl gnutls nss" would become e.g. |
| 8 |
> in the ebuild: |
| 9 |
> USE="ssl=[member-of: openssl,gnutls,nss]" |
| 10 |
> in make-conf: |
| 11 |
> USE: ssl=openssl, as synonymous of old USE="ssl openssl", or |
| 12 |
> USE: ssl=none, as synonymous of old USE="-ssl" |
| 13 |
> 2) subset-of: for selecting a number of modules to build. This would |
| 14 |
> replace USE_EXPAND quite neatly |
| 15 |
> 3) boolean: as alias for member-of: [true,false] |
| 16 |
> 4) unsigned int: IIRC some (few) packages can take optional uint at |
| 17 |
> configure-time. for example with dev-lisp/sbcl one can customize some |
| 18 |
> hardcoded GC parameters suchs as the default heap size, generation size, |
| 19 |
> etc... |
| 20 |
> |
| 21 |
> In the above case, one could have a USE variable named "webkit" of type |
| 22 |
> member-of: [qt,gtk] |
| 23 |
|
| 24 |
We can get similar results using booleans with REQUIRED_USE, and your |
| 25 |
approach seems to introduce needless complexity. |
| 26 |
-- |
| 27 |
Thanks, |
| 28 |
Zac |
Archives